OpenAI's GPT-Image-2 now removes backgrounds automatically

Populous, the architecture firm behind landmark venues like Wembley Stadium and the Las Vegas Sphere, has begun integrating Runway's generative video tools into its design workflow. The firm reports that the technology helps teams visualize large-scale venues before construction and has reclaimed roughly two weeks of time previously spent on competition pitches. It is a practical case study in how generative AI is finding a foothold in high-stakes architectural practice.

Meta's advertising platform ran ads promoting an app that claimed to generate non-consensual nude images of female politicians, with at least one ad featuring explicit deepfake content resembling a sitting US politician. The incident raises renewed questions about how effectively major platforms screen AI-enabled content that targets real individuals. It is the latest example of generative image tools being weaponized for harassment and political targeting.

Meta's Muse Video model has entered closed beta, offering early testers a look at its 10-second video generation capabilities. The model stands out for its native audio integration and reportedly strong temporal consistency across frames. These early outputs offer one of the first concrete glimpses into Meta's push into generative video.

A growing trend on X involves nonconsensual AI-generated images that have been subtly altered - tweaked just enough to evade automated detection while still being clearly identifiable as depicting real individuals. The technique is making moderation significantly harder and is spreading rapidly across the platform. 404 Media reports on how these so-called "subtlefakes" are changing the landscape of image-based abuse.

Eddie AI has announced a specialized 9-billion-parameter video editing model that trains privately for individual customers, keeping creative knowledge fully under their control. Built on a fine-tuned version of Qwen3.5-9B, the model achieves up to a 47% win rate against Kimi K3 on narrative arc identification - a result that puts it close to parity with a much larger general-purpose system. Its compact size also means it can run on standard GPUs and be deployed on-premises.

OpenAI has added transparent background support to its GPT-Image-2 model, available in preview through the API. Rather than removing a background after the fact, the model bakes the alpha channel directly into the generation process. The feature is enabled with a single parameter.